mainReduceNonCommutativeWGSL¶

Overview¶

A specialized raked reduce for when our input is non-commutative AND stored in a blocked (not striped) order. We essentially serialize some of it (reading a workgroup-size chunk at a time, reducing it, then loading the next).

@author Jonathan Olson <jonathan.olson@colorado.edu>

Type mainReduceNonCommutativeWGSLOptions¶

import type { mainReduceNonCommutativeWGSLOptions } from 'scenerystack/alpenglow';

• input: BufferSlot<T[]>
• output: BufferSlot<T[]>
• binaryOp: BinaryOp<T>
  TODO: length handling?!?
• stripeOutput?: boolean
  We can stripe the output (so the next layer of reduce can read it as striped)
• reduceOptions?: StrictOmit<reduceWGSLOptions<T>, "value" | "scratch" | "workgroupSize" | "binaryOp" | "localIndex" | "scratchPreloaded" | "valuePreloaded" | "mapScratchIndex" | "convergent">
  e.g. something in the future?
• & RakedSizable & OptionalLengthExpressionable

Source Code¶

See the source for mainReduceNonCommutativeWGSL.ts in the alpenglow repository.